In my lecture notes, I am given the definitions for:

-the independence of two discrete random variables

-the independence of a set of discrete random variables

-the pairwise independence of a set of discrete random variables

When learning about probability generating functions, we have the theorem:

Let $X_1, X_2, \ldots$, be i.i.d. non-negative integer-valued discrete random variable and let $N$ be another non-negative integer-valued discrete random variable, independent from $\{X_1, X_2,\ldots\}$ then the probability generating function of the random sum

$$X_1 + X_2 + X_3 + \cdots + X_N$$

is $G_N(G_X(s))$.

My question is, what does it mean for N to be 'independent from $\{X_1, X_2,\ldots\}$'? Does it simply mean the set $\{N, X_1, X_2,\ldots\}$ is independent?

This is an introductory course in probability, please don't give a measure theoretic explanation.


1 Answer 1


$N$ being independent from $\{X_1,X_2,\ldots\}$ does not mean that $\{N,X_1,\ldots,\}$ is independent. Use your intuition about what "independent" should mean. Does $N$ being independent from some things mean those things are independent of eachother? No that doesn't sound quite right. If $N$ is independent of $X_1$ and $X_2$, it definitely means $N$ is independent of $X_1$ and $N$ is independent of $X_2$. But sometimes this is not enough as the information of $X_1$ and $X_2$ together may somehow give more information than expected. For instance, if I flip a coin a random number $N$ times, get $H$ heads and $T$ tails, knowing $H$ doesn't tell me much, knowing $T$ doesn't tell me much, but knowing $H$ and $T$ also tells me $N$. To account for this kind of scenario we must carefully choose our definition of independence to make sure that $N$ is independent from joint distributions of the $X_i$ too.

The definition of $N$ being independent from $\{X_1,X_2,\ldots\}$ is any one of the following equivalent conditions:

1) for any $n \in \mathbb{N}$ set of distinct indices $i_1,\ldots,i_n$ and any $t_0, t_1,\ldots, t_n \in \mathbb{R}$ we have $$P(N \leq t_0, X_{i_1} \leq t_1, \ldots, X_{i_n} \leq t_n) = P(N\leq t_0) P(X_{i_1} \leq t_1, \ldots, X_{i_n} \leq t_n).$$

Note: we do not split $P(X_{i_1} \leq t_1, \ldots, X_{i_n} \leq t_n)$ into $P(X_{i_1} \leq t_1)\cdots P(X_{i_n} \leq t_n)$ unless we are also requiring all the $X_i$ to be independent of eachother.

2) for any $n \in \mathbb{N}$ and any set of distinct indices $i_1,\ldots,i_n$ and any intervals $I_0,\ldots,I_n$ we have

$$P(N \in I_0, X_{i_1} \in I_1, \ldots, X_{i_n} \in I_n) = P(N\in I_0) P(X_{i_1} \in I_1, \ldots, X_{i_n} \in I_n)$$

3) for any $n \in N$ and any bounded continuous $f:\mathbb{R}\to\mathbb{R},g:\mathbb{R}^n \to \mathbb{R}$ we have

$$ E[f(N)g(X_1,\ldots,X_n)] = E[f(N)] E[g(X_1,\ldots,X_n)] $$

and there are many other equivalent conditions. Their equivalence and the most natural definition of "independent" is more the subject of a measure theory and so I will not go into detail.

  • $\begingroup$ Is the definition similar for events? eg. If A and B are independent from X,Y,Z, do we mean P(A and B and X and Y and Z) = P(A and B) x P(X and Y and Z)? $\endgroup$ Mar 18, 2015 at 9:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.